phoenix-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "ChiaPing Tsai (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (PHOENIX-3569) Deadlock when updating both of local and global index
Date Tue, 10 Jan 2017 00:25:58 GMT

    [ https://issues.apache.org/jira/browse/PHOENIX-3569?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15813340#comment-15813340
] 

ChiaPing Tsai commented on PHOENIX-3569:
----------------------------------------

bq. Moving the index updates to postPut/postDelete will not ensure actual data and index consistency
The local index is updated in the posPut/postDelete, so is it also inconsistency before HBASE-15600?

bq. I will check with the test case to reproduce the issue.
Would you please take a look at the duplicate updates? The "testPostHook" in patch may provide
inspiration.

> Deadlock when updating both of local and global index
> -----------------------------------------------------
>
>                 Key: PHOENIX-3569
>                 URL: https://issues.apache.org/jira/browse/PHOENIX-3569
>             Project: Phoenix
>          Issue Type: Bug
>    Affects Versions: 4.9.0, 4.8.2
>            Reporter: ChiaPing Tsai
>            Priority: Critical
>         Attachments: PHOENIX-3569.fail.patch, PHOENIX-3569.v0.patch, PHOENIX-3569.v1.patch,
hbase-root-regionserver-node04.ncku.edu.tw.log
>
>
> We update the global index in Observer#postBatchMutate, but the hook is called before
completing the mvcc.
> If ParallelWriterIndexCommitter are busy to update the local index, the other updates
for global index will be blocked and the mvcc of data table won't be completed. The deadlock
happens since the ParallelWriterIndexCommitter will never complete the updates for local index.
> We can reproduce the deadlock through the following step.
> # create a data table with a local and a global index.
> # create 10+ (DEFAULT_CONCURRENT_INDEX_WRITER_THREADS) threads which disable the auto
commit to submit data
> # wait for the deadlock. the rs log will show the following message. "STUCK: MultiVersionConcurrencyControl{readPoint=8790,
writePoint=8801}"
> It seems to me that we should update the both of local and global index on the hook of
postPut and postDelete.
> Any suggestions are welcomed.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message